Hypothesis ranking (HR) is an approach for improving the accuracy of both domain detection and tracking in multi-domain, multi-turn dialogue systems. Language models for HR are typically language/locale dependent where recognition of input depends on a dialogue system being trained to understand a certain language. It is with respect to this general technical environment that the present application is directed.